organic solvent nanofiltration (osn): a new technology platform … · 2017. 7. 12. ·...
Post on 24-Mar-2021
0 Views
Preview:
TRANSCRIPT
1 | P a g e
Organic Solvent Nanofiltration (OSN): A New Technology Platform for Liquid-
Phase Oligonucleotide Synthesis (LPOS)
Jeong F. Kim1‡, Piers R. J. Gaffney
1‡, Irina B. Valtcheva1, Glynn Williams
2, Andrew M.
Buswell2, Mike S. Anson
2, Andrew G. Livingston
1*
1. Department of Chemical Engineering, Imperial College London, Exhibition Road,
London, SW7 2AZ.
2. GlaxoSmithKline Medicines Research Centre, Gunnels Wood Road, Stevenage, Herts,
SG1 2NY, UK. ‡. These authors contributed equally. *. Corresponding Author: a.livingston@imperial.ac.uk
2 | P a g e
3 | P a g e
Abstract
Organic Solvent Nanofiltration (OSN) technology is a membrane process for
molecular separation in harsh organic media. However, despite having well-documented
potential applications, development hurdles have hindered the widespread uptake of OSN
technology. One of the most promising areas of application is as an iterative synthesis
platform, for instance for oligonucleotides or peptides, where a thorough purification step
is required after each synthesis cycle, preferably in the same working solvent. In this
work, we report a process development study for liquid-phase oligonucleotide synthesis
(LPOS) using OSN technology. Oligonucleotide (oligo) based drugs are being advanced
as a new generation of therapeutics functioning at the protein expression level. Currently,
over one hundred oligo based drugs are undergoing clinical trials, suggesting that it will
soon be necessary to produce oligos at a scale of metric tons per year. However, there are
as yet no synthesis platforms that can manufacture oligos at >10 kg batch-scale. With the
process developed here, we have successfully carried out 8 iterative cycles of chain
extension and synthesized 5-mer and 9-mer 2’-O-methyl oligoribonucleotide
phosphorothioates, all in liquid phase media. This paper discusses the key challenges,
both anticipated and unexpected, faced during development of this process, and suggests
solutions to reduce the development period. An economic analysis has been carried out,
highlighting the potential competitiveness of the LPOS-OSN process, and the necessity
for a solvent recovery unit.
Keywords: Organic solvent nanofiltration, OSN, membrane, oligonucleotide, liquid
phase synthesis
4 | P a g e
1. Introduction The discovery of RNA interference (RNAi), and its exquisite control of protein
expression, has reinvigorated the consideration of oligonucleotide (oligo) based drugs as
potential therapeutic agents1-5
. There are currently three oligo-based drugs approved by
Food and Drug Administration (FDA): formivirsen (Vitravene), mipomersen (Kynamro),
and pegaptanib (Macugen). The first two operate through anti-sense mechanisms, while
the last is an aptamer. Following on from these successful market entries, over 100 oligo-
based drugs are now going through different phases of clinical trials6.
One of the major challenges for the exploitation of oligo-based therapeutics, apart
from the difficulties in drug delivery7, 8
, is the large-scale production of these complex
molecules. As more drugs are developed which target a broader range of patients, it is
anticipated that metric tons per annum of such drugs will soon be required9, 10
. Thus,
there is a pressing need to develop economically viable processes to produce oligos at
this scale 10-12
.
The chemical basis of the synthetic methodology has remained essentially the same
for the last 40 years. DNA and RNA are heteropolymers, with their mode of action
defined by the sequence of the monomers. Therefore the synthesis employs step-wise
addition of each nucleotide monomer, followed by separation of the reagent debris. The
current state-of-the-art oligo synthesis platform is solid-phase oligonucleotide synthesis
(SPOS) 6. The key feature of SPOS is the use of a solid support to which one end (the 3’-
end) of the growing chain is anchored. After immersion of the tethered oligo in a solution
of reactive monomer, the solid support facilitates rapid removal of the excess reagents
after each step, where the solid beads are simply washed with solvents. The growing
chain on the solid support is then ready for the next round of chain extension reaction. To
date, the versatile SPOS platform, with its high efficiency, has been able to meet the
small-scale demands of research and there have been no fundamental changes to the
SPOS process for the past 20 years.
Despite its clear advantages of easy purification and rapid synthesis, the general
consensus with regards to scaling up the SPOS process is that it is very difficult for
several reasons11, 13, 14
. First, the heterogeneous nature of the SPOS process leads to high
mass-transfer resistance between the bulk solution and the support surface, and
5 | P a g e
introduces steric hindrance from the support. Thus SPOS requires a higher excess of the
expensive monomers to drive coupling to completion than would be required in solution
phase. Secondly, the intensity of SPOS is restricted by sub-maximal loading of the
polymeric support to minimize interference between neighboring chains (particularly
when preparing longer oligos), and thus maintain coupling efficiency15
. Thirdly, the
SPOS process is highly mass intensive at large scale6. For instance, to produce 1kg of 20-
mer phosphorothioate oligos, approximately 4,000 kg of solvents, reagents, and water are
needed16
.
Another factor that is rarely considered in SPOS scale-up is the difficulty of
intermediate analysis. On a large scale, where the loss of one batch is highly undesirable
in terms of economics and operation time, sampling each cycle would be of great utility
to detect failed steps when remedial measures can still be effected. However,
representative sampling of solid phase from a large, packed-bed reactor is technically
challenging, and in practice is not performed. Thus, product assay is again conducted
only once the entire synthesis is complete, and in the event of one poor step the whole
batch must be extensively purified, or even discarded. Despite the drawbacks, many
significant breakthroughs have been made in SPOS scale-up recently, as reviewed by
Sanghvi6.
Liquid-phase oligonucleotide synthesis (LPOS) has long been proposed as an
attractive alternative large scale platform to SPOS17
. Unlike SPOS, the homogeneous
LPOS process does not suffer from mass-transfer limitations between the bulk solution
and the surface of the support, allowing lower reagent excess. LPOS also occupies a
lower reaction volume than SPOS because the inert interior of the solid support no longer
needs to be accommodated within the equipment. However, although LPOS offers
significant advantages over SPOS, it suffers the major drawback of the cumbersome
downstream purification required to remove reaction debris after each iterative cycle.
In LPOS, the growing oligo is often anchored to a soluble support which is used to
facilitate the downstream purification using techniques such as precipitation18, 19
,
extraction20
, or membrane separation17
. Bayer and Mutter17
first identified membrane
separation as a key adjunct to liquid phase peptide synthesis for downstream processing.
The same laboratory later applied this strategy to oligo synthesis using the obsolete
6 | P a g e
phosphodiester coupling strategy and diafiltering in aqueous solution21
. Until recently,
membranes could not effectively be employed for oligo purification because readily
available polymeric membranes were not stable in organic solvents and lacked molecular
discrimination. However, recent Organic Solvent Nanofiltration (OSN) membranes can
now withstand harsh conditions, such as aggressive solvents [e.g. N,N-
dimethylformamide (DMF) and N-methyl-2-pyrrolidone(NMP)] as well as the
oligonucleotide reaction solvent (acetonitrile), while being stable over a wide range of pH,
and resilient to both organic acids and bases 22
. More importantly, OSN membranes now
have nano-separation capabilities 23, 24
, discriminating solutes in the range between 100
and 2,000 Da.
Figure 1. Schematic of the LPOS-OSN process composed of 4 unit operations per cycle25
:
a) chain extension; b) first membrane diafiltration to wash out excess reagents; c)
deprotection of 5’-O using an acid to free the growing end; and d) second membrane
7 | P a g e
diafiltration to remove all the reagents except the growing oligos. [Reprinted with
permission from reference 25, Copyright Wiley-VCH Verlag GmbH & Co].
This research group has previously explored liquid-phase peptide synthesis26
, in
which the growing peptide chain was attached to a large mono-methoxy polyethylene
glycol (mPEG-5000) support. Filtration of crude reaction mixtures through an OSN
membrane allowed the excess reagents and impurities to pass, but the soluble peptide-
PEG conjugate was retained by the membrane, and was thus purified for the next step.
Building on this earlier experience, we have reported the application of OSN purification
to LPOS, henceforth a process termed LPOS-OSN25
. Using the LPOS-OSN process, we
have successfully synthesized 5-mer and 9-mer 2’-O-methyl phosphorothioate
oligoribonucleotides. The synthesis was carried out using commercially available 5’-O-
Dmtr-2’-O-methyl nucleoside phosphoramidite building blocks (Dmtr = 4,4’-
dimethoxytriphenylmethyl), employing the standard coupling method27
.
The entire process proceeds in the liquid phase, as illustrated in general schematic
Figure 1. A branched mono-disperse support was selected, on which three oligo chains
could grow simultaneously, increasing the atom efficiency of the process. The first
nucleotide monomer, N1, was attached to the soluble support via a succinate linker that is
cleaved at the end of the synthesis. To prepare an oligo of n non-identical monomers, the
iterative growth cycle then requires two synthetic steps, with membrane purification after
each, to be repeated up to the desired length: A building block (Nn-OP), bearing a
temporary protecting group (P), is first coupled to the chain terminus of the growing
oligo (Figure 1a); the temporary protecting group P is then removed (Figure 1c) to
commence the next chain extension cycle. After each reaction, the excess reagents are
washed through the OSN membrane (Figure 1b, d), in a process known as constant
volume diafiltration, CVD. Two specialized terms routinely employed in this paper are
retentate and permeate: These refer to the solutions retained by the membrane and
permeated by the membrane, respectively, during a CVD.
In this work, we report the development of an LPOS process using OSN as the key
separation technology. Typically, one of the main hurdles to the widespread applications
of membrane technology is the long and circuitous process development period. There
has been much effort in academia over the past decade to tap into the potential of OSN
8 | P a g e
technology, with some success. We have investigated some of the challenges encountered
during the process development stage, both expected and unexpected, and proposed
suitable solutions not only for this work, but for other potential OSN applications. In
addition, this first pilot project for LPOS-OSN was used to provide enough practical
experience to construct an economic analysis to assess the feasibility of scaling up the
proposed process. It is shown that the LPOS-OSN process is highly competitive and
scalable for large-scale oligo synthesis, and we assert that the LPOS-OSN process is a
promising platform to make large-scale (>100 kg) synthesis of oligonucleotides a reality.
2. Experimental
2.1. Materials
Acetonitrile (MeCN), methanol (MeOH), N,N-dimethylformamide (DMF),
dichloromethane (DCM), N,N-dimethylacetamide (DMAc), diethyl ether (Et2O), ethanol
(EtOH), and isopropyl alcohol (IPA), were HPLC grade from VWR, UK. Anhydrous
solvents were prepared by storing the solvents over baked 4Å molecular sieves cooled
under high vacuum (oil pump). Celazole® S26 polybenzimidazole (PBI) solution (26
wt% polymer in DMAc containing 1.5 wt% LiCl) was purchased from PBI Performance
Products Inc., USA. Non-woven polyolefin Novatexx 2471 was from Freudenberg
Filtration Technologies, Germany. The cross-linker α,α’-dibromo-p-xylene (DBX) was
purchased from VWR. 5’-O-(4,4’-Dimethoxytriphenylmethyl) (Dmtr) 2’-O-methyl
nucleoside phosphoramidite building blocks were purchased from ChemGene Corp., USA
or Fisher Scientific Ltd., UK. S-ethylthiotetrazole (ETT, 0.25M solution in MeCN,
Proligo), N-methylimidazole (NMI), phenylacetyl disulfide (PADS), pyridine, pyrrole
and dichloroacetic acid (DCA) were bought from Sigma-Aldrich, UK. Polyethylene
glycol (PEG) solutes were purchased from VWR, UK.
2.2 PBI 17DBX Membrane Fabrication & Testing Rig
Although several polymeric and ceramic membranes were screened for this work,
only the preparation of PBI membranes is described here, as they reproducibly gave the
desired performance (Please refer to Supporting Information for other membrane data).
9 | P a g e
PBI dope solutions were prepared by diluting the Celazole® S26 solution to 17 wt% with
DMAc. Once a homogeneous solution had been obtained, it was cast onto polyolefin
non-woven support using a casting knife (set to 250 µm). The supported film was quickly
immersed in a deionized water coagulation bath for phase inversion. The resultant
membrane was washed with IPA to ensure complete removal of water. To cross-link the
polymer, the membrane was immersed in a well-stirred 3 wt% solution of DBX in MeCN
and heated under reflux at 80 oC for 24 hours. The membrane was then removed from the
cross-linking solution and washed with IPA thoroughly. Finally, the membrane was
immersed in a PEG-400:IPA (1:1 v/v) impregnation bath for 4 hours, and then air-dried.
This formulation of PBI membrane is referred to as PBI 17DBX (17 wt% PBI cross-
linked with DBX). For further details of PBI membrane fabrication, readers are referred
elsewhere28, 29
.
For membrane testing and purification of oligos, membrane discs (effective area of
51 cm2) were loaded into four cells connected in series. The diagram of the process rig is
shown in Figure 2.
Figure 2. Membrane testing & diafiltration rig. Four membrane cells (effective
membrane area of 51 cm2 per cell) were connected in series (N = 4). The feed tank
(stainless steel) was pressurized using nitrogen, and the system volume was maintained
by constant addition of pure solvent back into the feed tank.
The feed tank was pressurized to the desired pressure (typically 10 bar) with nitrogen
and the gear pump (Michael Smith UK Ltd. GL Series) provided a flow rate of 66 L.hr-1
.
Before running preparative oligo purifications using PBI membranes, the membranes
10 | P a g e
were first characterized for rejection performance using a mixture of five linear PEGs
(MW of 200, 400, 1000, 2000, and 8000 Da), 1 g.L-1
of each dissolved in MeCN. For
preparative oligo purifications, the rig was charged with the crude mixture and the total
system volume was set to 0.4 L. The solvent was either neat MeCN, or MeCN mixed
with varying proportions of MeOH (discussed in detail in the Results section). To balance
the loss of system volume to the permeate, an equal volume of pure solvent was pumped
into the feed tank using an HPLC pump (Gilson 305, UK) to maintain a constant level.
The temperature of the rig remained constant at 21 ± 1 oC. Whenever necessary, samples
were taken from both the permeate and the feed tank for HPLC analysis.
In CVD, the term diavolume is a convenient time-like dimensionless variable defined
as the following
𝑑𝑖𝑎𝑣𝑜𝑙𝑢𝑚𝑒 =𝐽∙𝐴∙𝑡
𝑉𝑠𝑦𝑠𝑡𝑒𝑚 (Eq. 1)
where J represents the membrane flux (L.m-2
.hr-1
), A the membrane area (m2), t the
filtration time (hr), and Vsystem is the volume of the system (L). Simply put, one diavolume
means that the total permeated volume is equal to the system volume, which is
maintained at a constant level.
This parameter is useful for comparing CVD efficiencies and performances, and will
be used throughout this manuscript. For example, at the same final purity, a 10 diavolume
CVD is more solvent efficient than a 20 diavolume CVD.
2.3 Overall Process & Oligonucleotide Synthesis Chemistry
Figures 3 and 4 summarize the overall process and chemistry schemes, respectively,
for the LPOS-OSN platform developed in this work. The process is divided into three
phases: initial loading of the support, the chain extension cycle, and release and
purification of the final oligos. In the first phase, the first monomer of the chosen
sequence (2’-O-methyl uridine) was conjugated to the branched soluble support
[tris(octagol) homostar], after which the 5’-hydroxyl was deprotected to give mono-
nucleosidyl homostar using classical batch chemistry (see section 3.3). In the second
phase, the loaded homostar was passed through n-1 chain extension cycles until the
11 | P a g e
desired sequence was obtained (n-1 couplings are required to obtain an n-mer), using the
rig shown in Figure 2. In the last phase, the oligo was deprotected and cleaved from the
support, once more using classical solution phase chemistry, ready for final purification
by ion exchange chromatography.
In the above scheme the repetitive chain extension cycle is implemented using liquid
phase handling and membrane purification. This cycle consists of two synthetic steps: in
the first pair of reactions (coupling of the exposed 5’-OH to reactive phosphoramidite 5,
and thioylation of the internucleotidyl linkage with PADS) the next monomer is attached,
collectively known as chain extension; in the second step, known as detritylation, the 5’-
O-(4,4’-dimethoxytiphenylmethyl) (Dmtr, or dimethoxytrityl) protecting group is
removed from the 5’-terminus of the growing oligo ready for the next round of the cycle.
After the chain extension step the crude tris(Dmtr-oligonucleotidyl) homostar was
partially purified using membrane CVD to remove small molecules, and after
detritylation the crude tris(HO-oligonucleotidyl) homostar was again purified by
membrane CVD, this time removing all the monomer debris.
12 | P a g e
Figure 3. The LPOS-OSN platform: overall process flowchart. The process is split into
three phases: initial loading, chain extension cycle, and final purification. Each phase of
the protocol is detailed in the experimental section.
13 | P a g e
Figure 4. The LPOS-OSN platform: overall chemistry scheme for the same three phases.
The chemical synthetic procedures used in this study are described in full
elsewhere25
. The methods are summarized briefly below. The preparation of the
monodisperse octagol homostar 1 synthesis support was adapted from the literature
synthesis30
. The term homostar used in this manuscript refers to the branched polymer
support having three arms with a site on the end of each on which an oligo is grown.
Homostar 1 was first condensed with 5’-O-Dmtr-2’-O-methyl-3’-O-succinyl uridine 2
using 2,6-dichlorobenzoyl chloride (DcbCl) and N-methylimidazole (NMI), then the 5’-
hydroxyl was unblocked using dichloroacetic acid (DCA) and pyrrole to provide, after
chromatographic purification, loaded support 3.
Each chain extension reaction was slightly different due to the changing molecular
weight and solubility of each substrate, but a general chain extension protocol is
described below: The starting oligonucleotidyl homostar (ca. 1.2 g) was dissolved in a
small volume of DMF (2-4 mL) under N2 and co-evaporated from MeCN (3 × 20 mL).
The next phosphoramidite monomer (4.5 eq., i.e. 1.5 eq. per 5’-OH) was added as a dry
14 | P a g e
solid, and the mixture was dissolved in 0.25 M ETT in MeCN (9 eq.). After stirring for
35 min, PADS (9 eq., i.e. 3 eq. per 5’-OH) was added, the volume was doubled with
pyridine, and stirring continued for another 30 min. Notably, PADS was used as obtained,
instead of keeping a solution of PADS in pyridine or 3-picoline to develop overnight.
An additional capping step is usually included in the SPOS cycle, using acetic
anhydride plus N-methylimidazole (NMI) to block any unreacted hydroxyl groups
remaining after coupling as acetate esters, and so prevent them participating in
subsequent chain extension cycles. However, we were unable to detect any incomplete
reaction by HPLC, even though analysis of our homostar supported oligos should be
especially sensitive to incomplete coupling because of the 3-arm geometry. Since there is
no requirement for mass transfer between the bulk liquid and a solid phase interface,
there was reason to suspect that couplings to our fully dissolved supported oligo would be
faster and more complete than SPOS couplings (see Introduction). Hence, the capping
step was deliberately omitted to shorten the chain extension cycle.
After chain extension and the subsequent membrane CVD, the supported oligo was
washed out of the membrane rig and the solvent evaporated. The residue was re-
suspended in dry DCM to which pyrrole (2 vol%) then DCA (1 vol%) were added to
effect detritylation (deprotection of the Dmtr protection group); upon addition of DCA
the substrate dissolved and an intense orange color appeared which dissipated over the
next 30 min. To ensure complete removal of all the 5’-O-Dmtr ethers the reaction was
monitored by TLC, when partially deprotected intermediates appeared as a ladder of
spots that turned orange in trifluoroacetic acid vapor. If the reaction did not approach
completion in 15 min, a further aliquot of DCA was added; large oligos tend to buffer
acidic solutions, slowing detritylation. Upon completion of the reaction, the reaction was
quenched with a volume of pyridine equal to the total amount of DCA used
(approximately equi-molar amounts); with longer oligos a thick precipitate formed at this
point, presumed to be the product. The solids were re-dissolved when diluted into MeOH-
MeCN, and this solution was then subjected to another round of CVD.
To effect global deprotection and cleavage from the homostar support, a sample of
supported oligo was suspended in MeCN to which was added diethylamine (20 vol%).
15 | P a g e
After 30 min the mixture was filtered through a pad of cotton wool to remove any
insoluble material, the solvent was evaporated in vacuo, and the residue was re-dissolved
in conc. ammonia. The solution was heated at 55 C overnight and the next day filtered
once more. The filtrate was evaporated to dryness, and the residue co-evaporated from
ethanol (3×). Finally the crude solids were triturated with MeCN to remove protecting
group and support debris.
2.4 Membrane Purification – Constant Volume Diafiltration (CVD)
After the chain extension (section 2.3), the crude material was first diluted with
either MeCN or MeOH-MeCN (250ml, 5-20% MeOH, depending on the solubility of the
oligonucleotidyl homostar) and the rig (containing 150ml of the same solvent) was
charged with this solution. The rig was then pressurized to 10 bar to commence CVD,
and this continued until a total of 13-15 diavolumes had permeated, after which the
retentate was washed out of the rig; the solvent was evaporated to provide a mass balance,
and the residue analyzed by NMR, MS, and HPLC.
After detritylation (section 2.3), the crude suspension (in DCM, DCA, pyridine and
pyrrole) was diluted with MeCN or MeOH-MeCN (trinucleotide onwards; 250 ml, 5-20%
MeOH). To ensure that DCM did not damage the O-ring seals or membranes, the crude
solution was concentrated on a rotary evaporator until the vapor pressure reached 20
mbar at a bath temperature of 30 C, when DCM was assumed to have been removed and
further evaporation stopped. The crude solution was returned to the rig, which was once
more pressurized to 10 bar for a second round of CVD. This time, the solvent for the first
5 diavolumes contained 1 vol% pyridinium dichloroacetate (Py.DCA) in MeCN or
MeOH-MeCN (trinucleotide onward), followed by 10 diavolumes with neat solvent.
After CVD the purified retentate was washed out of the rig, and the solvent evaporated.
Finally, although it probably does not interfere in the subsequent chain extension reaction,
residual Dmtr-pyrrole was removed by precipitation of the oligonucleotidyl homostar
product into diethyl ether to provide an accurate mass balance, and the purified
tris(oligonucleotidyl) homostar was again analyzed by NMR, MS, and HPLC.
16 | P a g e
It had been hoped that it would be possible to carry out only one CVD per cycle by
detritylating immediately after chain extension. However this was not successful, and led
to many unidentified side products of longer HPLC retention time if left for any length of
time (Supporting Information). Accumulation of even low levels of such contaminants
with repeated cycles of chain extension could not be risked, which is why we have
adopted the double CVD approach reported here.
2.5. Analytical Methods
During the chain extension cycle, all reactions were monitored by high performance
liquid chromatography (HPLC). CVD purification was also followed by HPLC to
determine when the retentate could be removed from the rig. An Agilent 1100 Series
HPLC system was employed equipped with a UV detector and Varian 385-LC
evaporative light scattering detector (ELSD). The pump flow-rate was set at 1 ml.min-1
,
the injection volume was 30 µL, the column temperature was 30 C, and an ACE C18 RP
column (Hichrom Ltd, UK) was fitted. The column was eluted with a gradient of MeOH
and water buffered with 100 mM ammonium acetate. The UV wavelength was set at 260
nm, the ELSD evaporation temperature was set to 40 C, nebulization temperature at 55
C, and the nitrogen gas flow rate was at 1.5 SLM (standard liter per minute).
Nuclear magnetic resonance (NMR) spectra were recorded on Brüker AV-400 or
Brüker AV-500 spectrometers. Mass spectra (MS) were recorded on Micromass MALDI
micro MX, or Micromass LCT Premier (ESI) mass spectrometers.
17 | P a g e
3. Results & Discussion 3.1. Membrane Screening & Characterization
Several different types of membranes, including polyimide, polyamide thin film
composite (TFC), ceramic membranes, and a cross-linked polybenzimidazole (PBI)
membrane were screened for use in LPOS-OSN. The criteria for membrane selection
were: 1) chemical and solvent stability, 2) high rejection of the growing oligos (three i-
mers attached to the hub molecule, where i = 2 to n), 3) low rejection of the impurities
(excess monomers), and 4) durability and longevity. Polyimide and TFC membranes
were not stable for prolonged periods under the oligonucleotide synthesis conditions, as
shown in the work of Valtcheva et al.,22
and thus were discarded. Of the other types of
membrane, most did not reject the product highly enough while simultaneously allowing
the impurities to permeate through. However, PBI membranes cross-linked with DBX
(PBI 17DBX) exhibited promising PEG rejection data, shown in Figure 5. Furthermore,
the chemical stability of PBI membranes and their reproducible performance in solvent
environments had already been demonstrated in the work of Valtcheva et al.22
Please see
Supporting Information for full screening data.
PBI 17DBX membranes showed exceptional chemical stability towards all the
reagents used in this work, as well as reliable mechanical stability and long service life,
with excellent durability. It was determined that the threshold pressure of PBI 17DBX
membranes is approximately 15 bar, i.e. above 15 bar the membrane underwent
irreversible compaction and a permanent loss of membrane performance. Hence, the
membrane was operated at 5-10 bar at all times, which ensured constant membrane
performance for more than a year (See Supporting Information for long-term data). The
membrane permeance remained virtually constant at approximately 8 L.m-2
.hr-1
.bar-1
over
the entire duration of the 2’-O-methyl RNA phosphorothioate nonamer (9-mer) synthesis.
Slight variations in flux were observed, depending on the solution concentration and
solvent composition, but periodic testing of the PEG rejection by the membranes between
preparative experiments confirmed that the performance remained substantially
unchanged.
3.2 Justification for Using a Branched PEG Homostar Support
18 | P a g e
When performing solute fractionation operations using membranes, the constant
volume diafiltration (CVD) mode is usually employed, where the retentate volume is held
constant by matching the permeate volume outflow with pure solvent input. In CVD,
smaller solutes generally permeate through the membrane faster than bigger solutes: the
slower the permeation, the higher the rejection by the membrane, and vice versa (Eq. 2).
Consequently, the greater the difference in solute permeation rates (or rejections), the
easier the separation becomes. The term rejection is mathematically defined as the
following:
Ri(%) = [1 − (𝐶𝑃𝑖
𝐶𝑅𝑖)] × 100% (Eq. 2)
where CPi and CRi represent the concentrations of species i in the permeate and retentate,
respectively.
In this work the growing oligos [supported as tris(oligonucleotidyl) homostars] are
the products that must be highly retained by the membrane, whereas it is desirable that
everything else (excess building blocks and other reaction debris) should permeate
through. By writing a mass balance around the system (see Appendix A), it can be
shown that to achieve a clean and efficient separation between large and small solutes
using diafiltration, two conditions need to be met: 1) the difference in rejection between
the two solutes needs to be wide; and 2) the rejection of the larger compound should be as
close to 100% as possible. The first condition determines the efficiency of separation,
whereas the latter determines the yield of the process.
This is illustrated in Figure A1 (Appendix) where the normalized solute
concentration (C/Co) profile is plotted against the number of permeated diavolumes. It is
important to note that even with a seemingly high product rejection of 95%, the
remaining concentration (i.e. diafiltration yield) after 20 diavolumes is only 38%.
Therefore, to obtain a high diafiltration yield, the growing supported oligos must have
near to 100% rejection.
In the initial work the growing oligo was covalently attached to a long-chain linear
PEG support31
. PEG was selected as a polymer support because of its high solubility,
19 | P a g e
easy functionalization, and low cost. In addition, different molecular weights (MW) were
readily available. The PEG support increased the rejection of the oligo relative to the
debris, and widened the rejection difference between them, improving the discrimination
and efficiency of separation. However, it quickly became clear that it is difficult to obtain
rejections higher than 95%, regardless of the MW of the PEG. It was hypothesized that
because of the PEG’s highly flexible nature, it could thread, or reptate through the
membrane pores, lowering its rejection below that typically expected for more globular
molecules of a similar MW.
Figure 5. Comparison of linear versus branched PEG rejection by PBI 17DBX
membranes. Linear PEG rejections, regardless of their MW, do not exceed 95%.
Branched PEG compounds exhibit higher rejections and approach 100% above 3,000 Da.
Because of these initial observations, we investigated the potential of a branched
PEG synthesis support, as it has been shown that branched solutes exhibit higher
rejection than linear ones of comparable MW32
. The use of a branched support has the
additional benefit of increasing the loading of oligos per support molecule – from a
maximum of only two sites per linear PEG, to one site for every side-arm on a branched
support. 1,3,5-Tribromomethyl benzene was treated with a 10-fold excess of linear PEGs,
MWs ranging from 200 to 2,000 Da, to give crude mixtures containing approximately
20 | P a g e
seven parts starting linear PEG mixed plus one part 1,3,5-tris(PEG-oxymethyl)benzene.
The rejections of linear and branched, 3-armed PEG homostars by PBI 17DBX were
determined and are plotted against MW, Figure 5.
As can clearly be seen in Figure 5, whereas the rejection of the linear PEGs rose to
around 95%, but did not rise higher with growing molecular weight, above ca. 3,000 Da
the rejection of branched PEG homostars approached very closely to 100%. The
significance of this apparently small difference should not be understated, as it makes a
large difference to the overall yield after many diavolumes have permeated, as illustrated
in Figure A1. Thus it was anticipated that synthesizing oligos extending out from a
central branched synthesis support would much improve the yield over our initial
approach based on linear PEG supports.
Taking the homostar rejection effect as a general principle, we then realized that a
monodisperse PEG support for oligo synthesis would have significant advantages over a
more easily prepared poly-disperse support: its complete monodispersity would ensure a
discrete molecular entity, with a single MW (unlike a polydispersed support) facilitating
convenient analysis by mass spectrometry (MS), as well as NMR and HPLC. An octagol
homostar (1224 Da) was chosen as the monodisperse support30
for this work because
once the second residue (cytidine in the present work) is attached, the size of the
dinucleotidyl PEG homostar conjugate has already become large enough (2245 Da) to be
highly rejected by the PBI 17DBX membrane (>99%). In addition, as the oligo chains
grow, the rejection is expected to approach 100%.
3.3 Chain Extension & Detritylation (combining reactions and diafiltrations)
Using the procedure illustrated in Figures 3 and 4, we have successfully synthesized
2’-O-methyl RNA phosphorothioate pentamer (5-mer) and nonamer (9-mer), with the
sequences shown in Figure 4. The syntheses of 5-mer and 9-mer protected homostars
started from 0.7 g and 1.4 g of mono-nucleosidyl homostar 3, respectively. However, as
the oligos became longer, the reaction scale varied depending on the yield after each step,
and how much sample was retained for further analysis.
21 | P a g e
In this work the couplings employed a relatively low excess - 1.5 equivalents (eq. per
OH) – of phosphoramidite monomer 5. This compares favorably to more than 2 eq.
excess monomer typically employed in the reported SPOS strategies14
, although the exact
excesses used commercially are not publicly available. It should be stressed that in
oligonucleotide synthesis, the main economic driver is the phosphoramidite excess
(discussed in Section 3.5).
The post-reaction crude mixture contained multiple nucleotidyl byproducts. These
resulted from the conversion of the excess phosphoramidite monomer a mixture of four
different phosphoryl derivatives upon addition of PADS to the coupling mixture, as
shown in Figure 6. Apart from the monomer excess, the crude mixture also contained
ETT and its diisopropylammonium salt, pyridine, and a large number of PADS-related
species.
Figure 6. Monomer debris identified during the chain extension cycle. Thioamidate (6a,
Dmtr or H) and amidate (6b, Dmtr or H) are neutral species, whereas monothioate (6c,
Dmtr or H) and dithioate (6d, Dmtr or H) are charged species.
Because LPOS-OSN is a liquid phase method, it is straightforward to sample at any
stage of the process and to utilize versatile and highly informative analytical techniques
such as HPLC, NMR, and MS. HPLC can be used to determine if the reaction has
reached completion or not (See the Supporting Information), and also whether the
impurities have been completely removed. Since 31
P NMR only detects phosphorus-
bearing compounds, this tool is able to show what types of monomers were present, and
also whether they have been removed or not. MS was mainly used to confirm the identity
of the product, and for the dimer and trimer MS was also used to identify some minor
22 | P a g e
product-related impurities. The HPLC, 31
P NMR, and MS data of the dinucleotidyl
homostar (tritylated 7, and detritylated 8) are presented in Figure 7.
Figure 7. Analysis of a typical chain extension cycle, from mU loaded homostar 3 to 5’-
OH dinucleotidyl homostar 8. Each peak has been identified by comparison with pure
compound. a) HPLC of feed, retentate and permeate after chain extension, but before
detritylation. Small molecule reagents – PADS, ETT, and pyridine – permeate through
the PBI membrane, but monomer-related species are retained. b) HPLC after
detritylation; excess monomers permeate but Dmtr-pyrrole only permeates partially. c) 31
P NMR of retentates before and after detritylation. Three classes of monomer debris
(except thioamidate 6a) are present before detritylation, but are absent after detritylation.
d) MALDI-TOF+ MS confirming the presence of compound 7 and 8. Also, N-
23 | P a g e
deacetylation of cytosine base for 8, a common impurity, is identified ([8−Ac+2H]+ =
3495.6, calc. C142H212N18O71P3S3+
= 3495.2).
Figure 7 summarizes a typical set of data obtained for the first chain extension cycle;
although the HPLC chromatogram becomes broader for longer oligos (discussed below),
the 31
P NMR and MS data are also representative of later cycles. Figure 7a depicts the
HPLC traces of dinucleotidyl homostar 7 (5’-O-Dmtr) post-chain extension, both before
and after membrane CVD. All the HPLC peaks were resolved except for the thioamidate
(6a, R=Dmtr) peak, which overlapped with the product 7 (5’-O-Dmtr). PADS showed
more than one HPLC peak and their relative intensities varied with the batch. Figure 7a
demonstrates that all the small reaction debris, i.e. ETT, pyridine, and PADS, were
completely removed in the first diafiltration. However, all four types of excess monomer
debris (6, R=Dmtr) remained in the final retentate, as they were all highly rejected (>95%)
by the PBI membranes. The main reason for such high rejections was assumed to be the
bulky 5’-O-Dmtr protecting group.
Figure 7b shows HPLC traces for the same experiment, but post-detritylation of
dinucleotidyl homostar 8 (5’-OH), where the thioamidate (6a, R=H) and amidate (6b,
R=H) peaks overlapped with the product 8. After membrane CVD, only two peaks were
present in the final retentate: dinucleotidyl homostar 8 (5’-OH), and Dmtr-pyrrole.
Complementing the HPLC data, it can be seen in the 31
P NMR (Figure 7c) that after
detritylation all the charged thioate monomers (6c, 6d, R = H) were removed efficiently
by diafiltration. However, thioamidate (6a, R=H) and small traces of amidate (6b, R=H)
monomers were detected in the final retentate (thioamidate 6a not shown in Figure 7,
explained below). The measured rejection of amidate (6b, R=H) was ~85, but the
rejection of thioamidate (6a, R=H) was significantly higher at >99%. Thus, the amidate
(6b, R=H) was mostly removed by the second diafiltration, but the thioamidate (6a, R=H)
remained as a major component of the monomer debris.
Notably, the ratios of the concentrations of the four types of monomers varied
considerably from one chain extension to the next. Since the relative amounts of
monomer debris were only quantified (by 31
P NMR) after the first CVD, this variation
may be partially explained by slight differences in the rejections of the monomers.
24 | P a g e
However, it was found that reducing the excess of PADS from 10 to 3 equivalents per 5’-
hydroxyl consistently reduced the proportion of thioamidate generated to undetectable
levels. Thus, it was beneficial to use the lower excess of PADS to minimize the amount
of thioamidate generated, and so to maximize oligonucleotidyl homostar purity at the end
of each chain extension cycle. Figure 7d presents the MALDI MS data for the tritylated
and detritylated dinucleotidyl homostar (7 and 8, respectively), which clearly shows the
expected peaks of [7+Na+H2O]+
= 4484.1 and [8+H]+
= 3538.3.
The only significant contaminant remaining at this point, after the second CVD, was
Dmtr-pyrrole. Although this molecule does permeate PBI 17DBX (see permeate in lower
trace, Figure 7b), its absolute quantity and rejection was too high (>90%) to be removed
completely. Despite the relatively low MW of Dmtr-pyrrole, it was assumed that its bulky
shape, as well as a possible contribution from its hydrophobicity, made it difficult for this
molecule to permeate through the membrane. Although it probably does not interfere
with the next chain extension, in order to obtain accurate mass data we deliberately
precipitated the growing oligonucleotidyl homostar in diethyl ether to remove the
remaining Dmtr-pyrrole. Undoubtedly, this step should be avoided or replaced in a
scalable LPOS strategy, either by using a different membrane, or a different protecting
group.
3.2.2 Ion Exchange Transport through PBI membranes
After detritylation, it was initially observed that the rejection of thioate monomers
(6c, 6d, R=H) was over 95% in MeCN, making CVD impractical. Surprisingly, upon
addition of 1vol% DCA, the thioate monomers began to permeate. Unfortunately, the
high concentration of DCA caused significant capping of the 5’-OH in the crude tris(HO-
dinucleotidyl) homostar 8 as dichloroacetate esters, as well as acid catalyzed N-
deacetylation of the protected cytosine nucleobase. Assuming that thioate permeation
proceeded via an anion exchange mechanism, the DCA was later replaced by 1 vol%
pyridinium dichloroacetate (Py.DCA in 20% MeOH-MeCN), when it was again found
that the anionic thioate monomers (6c and 6d, R=H) permeated with rejections lower than
50%, but N-deacetylation was now very low. Taking into account that the thioate
monomers did not permeate at the tritylated stage (6c and 6d, R = Dmtr), this suggests
25 | P a g e
that their permeation occurs through a combination of both ion exchange and sieving
mechanisms. However, the higher rejections of neutral monomers (6a and 6b, R=H) were
unaffected. Considering that the difference in MW between the anionic thioate monomers
and neutral monomers is less than 84 Da, the differences in rejections were quite
remarkable.
From this behavior it was hypothesized that the anionic thioate monomers (6c, 6d),
paired with diisopropylammonium cations, permeated via an anion exchange mechanism.
Such a mechanism would be consistent with the poly-basic structure of cross-linked
PBI29
, which can deprotonate Py.DCA (PyH+.Cl2CHCO2
- ion pair). In addition, the PBI
backbone may contain some permanent positive charges on cationic imidazolium rings
that form during the membrane cross-linking step, shown in Figure 8. The addition of
Py.DCA could facilitate the movement of membrane-associated thioates (6a, 6b, R=H)
through the membrane by exchange of anions between the immobile imidazolium cations
and pyridinium cations in solution. A speculative transport mechanism is illustrated in
Figure 8.
Figure 8. Speculative transport of anionic thioate monomers (6a, 6b, R=H) through PBI
membranes via the ion-exchange mechanism. The cationic imidazole backbone allows
26 | P a g e
anionic species to exchange between the sites, pushing out the thioates to the permeate
side down the concentration gradient.
3.2.3. Synthesis of 9-mer
Figure 9. HPLC chromatograms up to 7-mer oligos (from 9-mer preparation). Tritylated
(-ODmtr) compounds have longer retention times than detritylated (-OH) species. The
peaks became too broad to be detected by HPLC after 7-mer.
From the outset, the choice of a mono-disperse support was seen as an opportunity to
exploit HPLC throughout oligo synthesis. Usefully, tritylated (-ODmtr) oligo-homostars
elute from the HPLC column later than detritylated (-OH) species, owing to the
hydrophobicity of the Dmtr protection. However, the peaks gradually broadened with
longer chain length, so that by 6-mer it was virtually impossible to detect and assay the
oligonucleotidyl homostar by HPLC (Figure 9). One of the reasons for this peak
broadening is thought to be the exponential increase in the number of diastereoisomers in
the oligonucleotidyl homostar’s phosphate backbone. For instance, the number of
diastereoisomers quadruples from 2-mer (four) to 3-mer (sixteen) (See the Supporting
Information). Furthermore, the solubility of the oligos in the HPLC mobile phase
27 | P a g e
(MeOH-water) decreased with the chain length, necessitating the use of more dilute
sample leading to weaker peak intensity. Hence, from 7-mer 17 onward we mainly relied
on the 31
P NMR for assessing the oligo purity. Nevertheless, HPLC has proved to be an
effective analytical tool to monitor the extension reaction, as well as to detect the
presence of impurities. In order to analyze the oligonucleotidyl homostars beyond 7-mer,
if no effective HPLC method can be found to monitor the intact oligonucleotidyl
homostars, a destructive method to remove the diastereoisomeric artifact (e.g. global
deprotection) could be employed.
3.2.3. Solubility of the Growing Oligonucleotidyl Homostar
The oligos’ solubility changed drastically with the growing chain length. It was
observed that their solubility in neat MeCN rapidly decreased with chain length, so that
while 2-mer 8 had moderate solubility in MeCN, 4-mer 12 was almost insoluble.
However, upon the addition of MeOH to the solution, the oligo solubility was markedly
enhanced (5-20% MeOH in MeCN). Notably, the oligos were insoluble in both neat
solvents, dissolving only in the mixture. The same was true with the chlorinated solvents,
where the growing oligos could be dissolved in mixtures of chloroform or
dichloromethane with methanol, but in neither alone. This suggests these species require
a protic solvent, possibly to disrupt hydrogen bonding of the nucleobases, to be solvated.
In addition, a higher fraction of MeOH was required with the increasing chain length,
possibly due to greater opportunities for intermolecular H-bonding with larger molecules.
Even so, the 9-mer 22 had excellent solubility (above 10g.L-1
) in 20% MeOH-MeCN.
Apart from the MeOH-MeCN mixtures, the oligonucleotidyl homostars also
exhibited excellent solubility in neat DMF (above 100 g.L-1
) and DMF-MeCN mixtures.
Even though oligonucleotide phosphoramidite couplings have been reported in DMF19
,
we noticed that the chain extension reaction was not as fast in neat DMF as in MeCN. As
the chemistry of the growing oligos is quite complex, it seems inevitable that a mixed
solvent will be required for both synthesis and purification. Nevertheless, a single-solvent
system for both reaction and separation is highly desirable from the process perspective
and will be explored further.
28 | P a g e
3.2.4. Yield of Oligonucleotide Synthesized by LPOS
Chain extension yields and the corresponding overall yield for 9-mer synthesis are
summarized in Figure 10 below. Yield values were calculated from the measured mass
data assuming 100% product purity, and are reported as molar yield.
Figure 10. Summary of chain extension yield and the corresponding overall yield,
starting from the mononucleosidyl homostar 3 as the basis of calculation: The yield is
calculated from the mass data and reported as molar yield, assuming 100% purity.
It was anticipated that bigger oligonucleotidyl homostars would exhibit higher
rejections by the membrane. In Figure 10 it can be seen that the step-wise yield improved
with the chain length as expected. Indeed, from the 5-mer onward the average step-wise
yield was higher than 95%. It can also be seen that the major loss in the overall yield
resulted from the first three chain extension cycles, contributing half of the total loss.
Although the rejection of the oligonucleotidyl homostar at the dimer stage was already
high at 99%, the isolated yield after 25 diavolumes was only 75%, as predicted from
Figure A1 (Appendix). The rejection of the oligonucleotidyl homostars rose to 99.8%
after tetramer. However, it should be noted that the calculated yield was solely based on
the mass data and the purity was not taken into account (only determined after global
deprotection). Hence, the apparent observed yield measurement must be higher than the
1 2 3 4 5 6 7 8 9 10
0
25
50
75
100
Oligonucleotidyl Homostar Size
Yie
ld (
%)
Step-wise Yield
Overall Yield
29 | P a g e
actual yield of pure oligo. Nevertheless, the rejection of the product is approximately
correct based on the HPLC integration, and the obtained yield values are likely to be
within errors.
It is often difficult to obtain a complete rejection (100%) using membranes, and
rejection values between 95 – 99.9% are commonly observed even for very tight
membranes33
. This is due to a low level of defects that inevitably arise during the
fabrication process, and/or to the permeation mechanism through the membranes.
Generally, the defect-related drop in rejection is eliminated when membrane modules are
employed, due to the much larger area to system volume ratio. Nevertheless, it is
anticipated that the maximum yield per chain extension is likely to be below 100% due to
incomplete rejection. To compensate for the significant losses in the first three chain
extension cycles, an alternative membrane configuration could be applied, such as a
membrane cascade34, 35
, which has been shown to increase the process yield without
compromising the purity.
3.4. Final deprotection and Purity Analysis
30 | P a g e
Figure 11. Final purity analysis by HPLC after global deprotection: a) 5-mer oligo
prepared by LPOS-OSN using 1vol% DCA; and b) 5-mer oligo by LPOS-OSN using
1vol% Py.DCA; c) 9-mer by LPOS-OSN, purified by ion exchange (49% before
purification); d) 9-mer by SPOS after desalting. Purity is roughly estimated by integrating
the product peak.
Two different batches of pentanucleotidyl homostar 14, and one batch of 9-mer
homostar 22, were cleaved from the core octagol-homostar support and analyzed for
purity by HPLC (Figure 11). For comparison, a 9-mer oligo sample of the same sequence
was also synthesized using a typical automated SPOS protocol. The main peaks are
indeed the desired 5-mer (23) and 9-mer (24) oligos; the purity was assessed by
comparing the integral ratio of the product peak to the other minor peaks. It can be seen
that the 5-mer purity was improved by switching from using DCA during OSN of the
detritylated oligo-homostars (66%) to Py.DCA (75%), presumably because capping of
the 5’-OH was prevented, consequently reducing the amount of short-mer contaminants
31 | P a g e
(n-1, n-2, …and/or n-i) visible at shorter retention times. The peaks after the product
suggest the presence of long-mer (n+i) impurities, which may be generated during chain
extension reaction when a newly coupled Dmtr-nucleotide was detritylated by the mildly
acidic ETT activator, leading to double chain extension. For this study we had employed
long coupling reaction times (>30 min) to allow time for concurrent HPLC analysis. It is
likely that many of these longer retention time contaminants can be largely eliminated by
reducing the coupling times to a more typical 3-6 minutes.
In general, the purities of oligos prepared by LPOS were not as high as those
prepared via SPOS. Compared to the impurity profile of SPOS derived oligos (Figure 11d)
where short-mers and long-mers peaks are clearly identifiable, the impurity profile of the
LPOS-prepared oligos (Figure 11a-c) showed many irregular and unidentified peaks that
do not overlap with the SPOS contaminants. Thus, a simple HPLC purity comparison
may not be precise. In the future, LC-MS techniques will be employed to understand this
impurity profile.
Nevertheless, the final purity of 9-mer was lower than anticipated (49% before
chromatographic purification), mainly because the last coupling reaction had not reached
completion (17% of n-1 impurity observed, confirmed by LC-MS). It is possible that the
chain extension did not reach completion because the 5’-mGmG dinucleotide is the most
hindered inter-nucleotide linkage to form. Even so, it can be seen that upon purification
of the 9-mer using ion exchange chromatography, the purity can be increased up to
acceptable standards (> 90%)6.
3.5. Economic Analysis
Currently, the major impediments for SPOS scale-up are the associated costs, and the
need for larger SPOS synthesizers. For instance, the cost of a 1kg scale synthesizer is
approximately £2 million. Also, the price of oligo therapeutics prepared by SPOS varies
significantly depending on the synthetic routes and chemical modifications. On the other
hand, the chief advantage of membrane-based processes is that they can be implemented
on almost any scale. To explore the competitiveness of the LPOS-OSN platform, an
32 | P a g e
economic evaluation of the proposed process was carried out from the data obtained in
this work.
The basis for calculation was the synthesis of 1 kg of 23-mer oligoribonucleotide.
The required reagents and solvents were back-calculated accordingly. Some necessary
assumptions were made: (i) initial capital investment on the membrane rig of £250,000
was amortized with a capital charge factor of 0.2; (ii) a total of 20 batches per year was
assumed (based on the operation time and membrane permeability of 8 L.m-2
.hr-1
.bar-1
);
(iii) the system volume was fixed at 100L (based on the required concentration); (iv) only
acetonitrile was used as the solvent; and (v) the labor and operation costs were not taken
into account. Although the membrane process is considered a low labor-intensity
process36
, an exact operation cost was difficult to estimate at this stage. It was found that
solvent accounts for a large proportion of the overall cost, and hence two different
scenarios were considered: the case with solvent disposal, and a scenario with 90%
solvent recovery. The calculated cost of solvent recovery (using steam for distillation)
contributes to 4% of the overall solvent cost. The overall cost breakdown is shown in
Table 1 and Figure 12 below.
Table 1. Overall cost breakdown per 1kg batch of oligo
Item Calculations Solvent Disposal Solvent Recovery
Capital Cost Amortize by 0.2, 20 batches.yr-1
£2,500 £2,500
Phosphoramidite Monomer
1.5 eq., £10.2.g-1 £53,000 £53,000
PADS, ETT, DCA, etc £1,500 £1,500
Solvent Acetonitrile, £0.8.L-1 £60,700 £6,300
Membrane 24m2, £1600.m-2 10 batches lifetime
£3,800 £3,800
Homostar support £100 £100
TOTAL (kg-1) £121,600 £67,200
TOTAL (g-1) £122 £67
33 | P a g e
Figure 12. Overall breakdown of the cost: (left) with solvent disposal and (right) with 90%
solvent recovery. The two largest fractions are the solvent and phosphoramidite monomer
costs. Implementing a distillation solvent recovery unit significantly reduces the overall
cost.
For the purpose of the calculation, the overall yield of 23-mer was assumed to be 50%
(step-wise yield of 97%). In the first case of solvent disposal, it is quite clear from Table
1 and Figure 12 that the two main cost components are the phosphoramidite monomers
and the solvent; other parameters, such as capital cost and membrane modules, are
relatively insignificant. The cost of phosphoramidite monomers has come down
significantly in the past 10 years in the face of rising demand37, 38
, but it is still a major
obstacle for commercial oligo production. In addition, the cost of phosphoramidite
depends significantly on the scale.
However, it can be seen that the overall cost of solvent when it is not recovered is
even higher than the phosphoramidite monomer itself. This is not surprising considering
that membrane CVD is a highly solvent-intensive process39
and acetonitrile is one of the
most expensive common solvents. For this reason, several approaches have been reported
that couple solvent recovery to membrane CVD35, 36, 40
. For the purpose of this calculation,
we have assumed 90% solvent recovery via distillation and have implemented the
associated energy cost. Distillation was chosen as a suitable solvent recovery unit as the
only other significant volatile compound employed in this project was pyridine. If 90%
solvent recovery is now taken into account, the cost of the oligo drops by 50% and the
biggest fraction of cost then becomes the phosphoramidite monomers.
34 | P a g e
It should be stressed that the labor and operation costs have not been taken into
account. Nevertheless, it can be asserted that the proposed LPOS-OSN platform is not
only highly competitive with the SPOS platform, but also more easily scalable. To
understand which operation parameters affect the overall economics, different sensitivity
analyses were carried out, as shown in Figure 13.
Figure 13. Sensitivity analyses aiming to understand how changes in key parameters
affect the overall cost: a) effect of monomer excess; b) effect of overall yield; c) effect of
membrane area; and d) effect of monomer cost.
Several trends can be observed in Figure 13. Firstly, for all cases the recovery of
solvent is economically beneficial, and environmentally desirable. Secondly, Figure 13a
demonstrates that there is a linear relationship between the molar excess of monomer and
the final oligo cost. For instance, decreasing the monomer excess from 1.5 to 1.1 eq. can
reduce the final product cost by up to 21%, something that is possible for an LPOS
platform where the reaction proceed in a homogeneous liquid phase, and mass transfer
does not limit reaction rates, thus allowing a lower reagent excess. Thirdly, the overall
yield has a steep inverse exponential relationship with the final cost (Figure 13b), and a
steeper gradient for the solvent-recovery case. For instance, increasing the overall yield
35 | P a g e
from 50% to 100% decreases the required monomer amount by 42%, and decreases the
overall cost by 27%.
Beyond these major considerations, it is very important to optimize the total
operation time with respect to the membrane area (Figure 13c). Because the membrane
area can easily be increased by using additional membrane modules, and the operation
time drops with higher membrane area, the operation time is an independent variable
under control. It should be stressed that the cost of membranes is only a small fraction (3-
6%) of the overall breakdown of the cost (Figure 12). It is expected that the membrane
area will be chosen based on the required campaign schedule considering the number of
batches, reaction time, and available reactor schedule, etc. The main cost optimization
parameters would be the overall yield and the monomer excess. Lastly, as expected, the
price depends significantly on the monomer price. If the monomer cost can be reduced
down from £10.2.g-1
to £3.5.g-1
, with increasing batch scale, the final product cost drops
by 52%.
3.6. Further Discussions: advantages and current challenges
The two batches of oligos synthetized here using the LPOS-OSN process clearly
demonstrated the expected advantages, such as efficient reaction, minimal usage of
phosphoramidite monomers, simple purification methodology, easy scalability, and in
situ analysis. However, several challenges have been identified during the development
stage, and the process needs to go through continuous improvement and optimization.
The current challenges are: 1) incomplete removal of excess monomers, especially the
amidate 6b, leading to lower purity; 2) low process yield; 3) incomplete removal of
Dmtr-pyrrole requiring a separate precipitation step; and 4) redundant solvent exchange
steps.
As for the first challenge, the excess phosphoramidite monomers are converted to
four different types of debris upon addition of PADS (Figure 6). By understanding the
mechanism of their formation, the reaction should be pushed towards the thioates (6c, 6d),
which have been shown to permeate more easily than amidates (6a, 6b) through the PBI
membranes, probably via an ion exchange mechanism. The second challenge, low
36 | P a g e
process yield, can be tackled through different membrane configurations. For example,
recent work by our group showed that employing a two-stage cascade can increase the
yield without compromising the product purity34, 35
. In addition, OSN membranes have
also shown potential for an in-situ solvent recovery 39
. Also, further improvements in
yield and purity can be made by parallel tuning of the size of the soluble support and the
membrane separation performance, to make the purification more efficient. Notably,
monodisperse PEG homostars have been prepared up to 8,000 Da 30, 41
, and the
membrane separation performance can be roughly controlled using polymer fraction in
the dope and the solvent to co-solvent ratio 42
. It should be noted that even though a
bigger soluble support may exhibit higher rejection, it also decreases the oligo loading
(mmols.g-1
of PEG support) while increasing the solution viscosity.
The use of Dmtr protecting group has been extremely versatile and useful for the
SPOS process, but it has proven to be too bulky for the proposed LPOS-OSN process, at
least with the PBI 17DBX membrane, requiring a precipitation step after every chain
extension cycle. Instead, other acid-labile protecting groups should be explored. For
example, the methoxy isopropyl acetal protecting group (MW 74) employed by Molina et
al.43, 44
to synthesize 3-mer oligos on a cyclodextran support could be applied.
Lastly, the current chain extension protocol requires redundant solvent exchanges
and a precipitation step, rendering the practical operation of the process cumbersome and
time-consuming. From the process operation perspective, a single-solvent and single-pot
synthesis protocol is preferred. One way to achieve significant process intensification is
to bypass the detritylation reaction in DCM, and perform the deprotection reaction and
membrane CVD in the rig (the membrane reactor concept). This is expected to
significantly simplify the process, as the entire chain extension cycle can proceed with
effectively a single CVD using a single-solvent system. All of these challenges are
currently under investigation and will be covered in our future publications.
Apart from the practical challenges identified in this work, the LPOS-OSN concept
still needs to prove its versatility and that it can produce other species apart from
antisense oligos, for instance spiegelmers, and locked nucleic acid (LNA), among other
types of oligo-based drugs.
37 | P a g e
4. Conclusions In this work we have introduced a new liquid-phase oligonucleotide synthesis (LPOS)
platform using organic solvent nanofiltration (OSN) as a scalable purification technology.
By growing oligos on a soluble, branched monodispersed PEG support, purification could
be performed after each reaction using OSN membranes, exploiting the difference in size
between the product oligonucleotidyl homostar and smaller reagent debris. Using this
new platform, 5-mer and 9-mer 2’-O-methoxy phosphorothioate oligoribonucleotides
have been successfully synthesized and characterized. The solubility of the
oligonucleotidyl homostars was excellent in mixed MeOH-MeCN solvent, as well as in
DMF. The synthesis was carried out on a gram-scale, which could easily be scaled-up to
kilo-scale using membrane modules. The main advantage of the proposed process is that
its performance is not affected by the scale of synthesis. We have chosen a PBI
membrane as it gave reliable and robust performance for over a year. A thorough
economic analysis was carried out for the LPOS-OSN platform to understand the main
cost factors, as well as operation parameters.
Although this work has shown the potential of the new liquid-phase platform for
oligo synthesis, continuous improvements and optimization steps must be taken. Several
process challenges have been identified, and possible solutions suggested. The proposed
LPOS-OSN platform presents a flexible alternative for the large-scale synthesis of oligos,
and allows economical synthesis of oligo-based therapeutics on a metric-ton scale. With
the ever-growing number of oligo-based drugs going through clinical trials, a scalable
manufacturing technology is expected to further boost growth in this exciting new era of
oligo therapeutics.
List of Abbreviations
Term Full name Term Full name
Cne cyanoethyl protecting group MeCN acetonitrile
CVD Constant Volume Diafiltration MeOH methanol
38 | P a g e
Da Daltons [g.mol-1] MS Mass Spectrometry
DBX dibromo-p-xylene MW Molecular weight
DCA dichloroacetic acid NMI N-methylimidazole
DcbCl 2,6-dichlorobenzoyl chloride NMP N-methyl-2-pyrrolidone
DMAc N,N-dimethylacetamide NMR Nuclear magnetic resonance
DMF N,N-dimethylformamide Nn nth nucleotide
Dmtr 4,4’-dimethoxytriphenylmethyl OSN Organic Solvent
Nanofiltration
DNA deoxyribonucleic acid P Protecting group
ELSD Evaporative light scattering
detector PADS phenylacetyl disulfide
ETT S-ethylthiotetrazole PBI polybenzimidazole
FDA Food and Drug Administration PEG polyethylene glycol
HPLC High pressure liquid
chromatography Py.DCA pyridinium dichloroacetate
IPA isopropyl alcohol RNA ribonucleic acid
J flux [L.m-2.hr-1] RNAi RNA interference
LNA locked nucleic acid SPOS Solid-phase Oligo Synthesis
LPOS Liquid-phase oligo synthesis TFC thin film composite
TLC thin layer chromatography
5. Appendix A: Process Modeling A constant volume diafiltration (CVD) system can be modeled by writing a mass balance
around the system. Assuming that the system operates at a constant volume and it is
perfectly mixed,
VdCR,i
dt= −F ∙ Cp,i = Jv ∙ A ∙ Cp,i (Eq. A1)
where V (dm3) is the entire system volume, F is the permeate flow-rate (dm
3.hr
-1), Jv
(dm3.m
-2.hr
-1) is the membrane flux, A (m
2) is the membrane area, and CR,i and CP,i
(g.dm-3
) are the concentrations of species i in the retentate and permeate, respectively.
39 | P a g e
Defining the observed rejection of species i as,
Robs = 1 −Cp,i
CR,i (Eq. A2)
substituting Eq. (A2) into Eq. (A1) yields,
dCR,i
dt= − (
1
V) JvA CR,i(1 − R𝑜𝑏𝑠) (Eq. A3)
This equation can be solved either numerically or analytically. When integrated
analytically with appropriate boundary conditions, the following equation is obtained:
Cr,i
𝐶𝑟,𝑖,𝑜= exp [−
𝐽𝑣𝐴𝑡
𝑉 (1 − 𝑅𝑜𝑏𝑠)] = exp[−𝑑𝑖𝑎𝑣𝑜𝑙𝑢𝑚𝑒 (1 − 𝑅𝑜𝑏𝑠)]
(Eq. A4)
where diavolume represent the total volume of permeate collected relative to the initial
system volume. This useful time-like dimensionless parameter allows different
diafiltration systems to be compared.
Figure A1. The effect of rejection on the overall concentration profile (i.e. diafiltration
yield versus diavolumes at different solute rejections).
0 5 10 15 200
20
40
60
80
100
Norm
. C
oncentr
ation C
/Co [
%]
Number of Diavolumes [-]
R = 100%
R = 99.5%
R = 99%
R = 95%
R = 90%
R = 50%
R = 25%
40 | P a g e
6. Acknowledgement The authors wish to acknowledge the following supports: J.F.K. and I.B.V. thank the
7th
framework program of the European commission’s Marie Curie initiative PITN-GA-
2008-238291-MEMTIDE; P.R.J.G. thanks the EPSRC for a Platform grant EP/J014974/1
and GSK.
7. Description of the Supporting Information
The supporting information file for this work includes: (1) effect of adding acid to
crude reaction, without initial diafiltration; (2) HPLC peak broadening effect with oligo
length; (3) effect of membrane cascade on overall process yield; (4) detection of
incomplete reaction using HPLC; (5) detailed preparation and screening data for tested
membranes; (6) long term performance data of PBI membranes.
41 | P a g e
8. References 1. Lebedeva, I.; Benimetskaya, L.; Stein, C.; Vilenchik, M., Cellular delivery of antisense
oligonucleotides. Eur. J. Pharm. Biopharm. 2000, 50, 101-119.
2. Fire, A.; Xu, S. Q.; Montgomery, M. K.; Kostas, S. A.; Driver, S. E.; Mello, C. C., Potent and specific
genetic interference by double-stranded RNA in Caenorhabditis elegans. Nature 1998, 391, 806-811.
3. Stein, C.; Cheng, Y., Antisense oligonucleotides as therapeutic agents--is the bullet really magical?
Science 1993, 261, 1004.
4. Fichou, Y.; Férec, C., The potential of oligonucleotides for therapeutic applications. Trends
Biotechnol. 2006, 24, 563-570.
5. Vaishnaw, A. K.; Gollob, J.; Gamba-Vitalo, C.; Hutabarat, R.; Sah, D.; Meyers, R.; de Fougerolles,
T.; Maraganore, J., A status report on RNAi therapeutics. Silence 2010, 1, 1-13.
6. Sanghvi, Y. S., A status update of modified oligonucleotides for chemotherapeutics applications. Curr.
Protoc. Nucleic Acid Chem. 2011, 4.1. 1-4.1. 22.
7. Branch, A. D., A good antisense molecule is hard to find. Trends Biochem. Sci. 1998, 23, 45-50.
8. Maier, M. A.; Jayaraman, M.; Matsuda, S.; Liu, J.; Barros, S.; Querbes, W.; Tam, Y. K.; Ansell, S.
M.; Kumar, V.; Qin, J., Biodegradable lipids enabling rapidly eliminated lipid nanoparticles for systemic
delivery of RNAi therapeutics. Mol. Ther. 2013, 21, 1570-1578.
9. Lajmi, A. R.; Schwartz, L.; Sanghvi, Y. S., Membrane purification of an antisense oligonucleotide.
Org. Process Res. Dev. 2004, 8, 651-657.
10. Reese, C. B.; Song, Q., The H-phosphonate approach to the solution phase synthesis of linear and
cyclic oligoribonucleotides. Nucleic Acids Res. 1999, 27, 963.
11. Andersson, L.; Blomberg, L.; Flegel, M.; Lepsa, L.; Nilsson, B.; Verlander, M., Large scale synthesis
of peptides. Peptide Science 2000, 55, 227-250.
12. Adamo, I.; Dueymes, C.; Schönberger, A.; Navarro, A. E.; Meyer, A.; Lange, M.; Imbach, J. L.; Link,
F.; Morvan, F.; Vasseur, J. J., Solution Phase Synthesis of Phosphorothioate Oligonucleotides Using a
Solid Supported Acyl Chloride with H Phosphonate Chemistry. Eur. J. Org. Chem. 2006, 436-448.
13. Bonora, G.; Rossin, R.; Zaramella, S.; Cole, D.; Eleuteri, A.; Ravikumar, V., A liquid-phase process
suitable for large-scale synthesis of phosphorothioate oligonucleotides. Org. Process Res. Dev. 2000, 4,
225-231.
14. Chen, C.-H.; Chen, W.-Y.; Chen, Y.-C.; Lee, M.-J.; Huang, C.-D.; Chanda, K.; Sun, C.-M.,
Convergent Solution Phase Synthesis of Chimeric Oligonucleotides by a 2+2 and 3+3 Phosphoramidite
Strategy. Aust. J. Chem. 2010, 63, 227-235.
15. So, S.; Peeva, L. G.; Tate, E. W.; Leatherbarrow, R. J.; Livingston, A. G., Organic Solvent
Nanofiltration: A New Paradigm in Peptide Synthesis. Org. Process Res. Dev. 2010, 14, 1313-1325.
16. Tedebark, U.; Scozzari, A.; Werbitzky, O.; Capaldi, D.; Holmberg, L., Industrial-Scale Manufacturing
of a Possible Oligonucleotide Cargo CPP-Based Drug. Methods Mol. Biol. 2011, 683, 505-524.
17. Bayer, E.; Mutter, M., Liquid phase synthesis of peptides. Nature (London, U. K.) 1972, 237, 512-513.
18. Bonora, G. M.; Scremin, C. L.; Colonna, F. P.; Garbesi, A., HELP (high efficiency liquid phase) new
oligonucleotide synthesis on soluble polymeric support. Nucleic Acids Res. 1990, 18, 3155.
19. Kungurtsev, V.; Laakkonen, J.; Molina, A. G.; Virta, P., Solution-Phase Synthesis of Short Oligo-2′-
deoxyribonucleotides by Using Clustered Nucleosides as a Soluble Support. Eur. J. Org. Chem. 2013,
6687-6693.
20. de Koning, M. C.; Ghisaidoobe, A. B. T.; Duynstee, H. I.; Ten Kortenaar, P. B. W.; Filippov, D. V.;
van der Marel, G. A., Simple and efficient solution-phase synthesis of oligonucleotides using extractive
work-up. Org. Process Res. Dev. 2006, 10, 1238-1245.
21. Brandstetter, F.; Schott, H.; Bayer, E., Liquid-phase-synthese von nucleotiden. Tetrahedron Lett. 1973,
14, 2997-3000.
22. Valtcheva, I.; Kumbharkar, S. C.; Kim, J. F.; Bhole, Y.; Livingston, A. G., Beyond polyimide:
Crosslinked polybenzimidazole membranes for organic solvent nanofiltration (OSN) in harsh environments.
J. Membr. Sci. 2014.
23. Vandezande, P.; Gevers, L. E. M.; Vankelecom, I. F. J., Solvent resistant nanofiltration: separating on
a molecular level. Chem. Soc. Rev. 2008, 37, 365-405.
42 | P a g e
24. Marchetti, P.; Jimenez Solomon, M. F.; Szekely, G.; Livingston, A. G., Molecular separation with
organic solvent nanofiltration: a critical review. Chem. Rev. (Washington, DC, U. S.) 2014, 114, 10735-
10806.
25. Gaffney, P. R. J.; Kim, J. F.; Valtcheva, I. B.; Williams, G. D.; Anson, M. S.; Buswell, A. M.;
Livingston, A. G., Liquid-Phase Synthesis of 2′-Methyl-RNA on a Homostar Support through Organic-
Solvent Nanofiltration. Chemistry – A European Journal 2015, 21, 9535-9543.
26. So, S.; Peeva, L. G.; Tate, E. W.; Leatherbarrowb, R. J.; Livingston, A. G., Membrane enhanced
peptide synthesis. Chem. Commun. (Cambridge, U. K.) 2010, 46, 2808-2810.
27. Beaucage, S., Solid-phase synthesis of siRNA oligonucleotides. Curr. Opin. Drug Discovery Dev.
2008, 11, 203.
28. Valtcheva, I. B.; Marchetti, P.; Livingston, A. G., Crosslinked polybenzimidazole membranes for
organic solvent nanofiltration (OSN): Analysis of crosslinking reaction mechanism and effects of reaction
parameters. J. Membr. Sci. 2015, 493, 568-579.
29. Valtcheva, I. B.; Kumbharkar, S. C.; Kim, J. F.; Bhole, Y.; Livingston, A. G., Beyond polyimide:
Crosslinked polybenzimidazole membranes for organic solvent nanofiltration (OSN) in harsh environments.
J. Membr. Sci. 2014, 457, 62-72.
30. Székely, G.; Schaepertoens, M.; Gaffney, P. R.; Livingston, A. G., Iterative synthesis of monodisperse
PEG homostars and linear heterobifunctional PEG. Polym. Chem. 2014.
31. Vasconcelos, R. C. Organic Solvent Nanofiltration in the synthesis of DNA oligonucleotides and
heterobifunctional polymers. Ph.D Thesis, Imperial College London, 2010.
32. Zheng, F.; Li, C.; Yuan, Q.; Vriesekoop, F., Influence of molecular shape on the retention of small
molecules by solvent resistant nanofiltration (SRNF) membranes: A suitable molecular size parameter. J.
Membr. Sci. 2008, 318, 114-122.
33. Baker, R. W., Membrane technology and applications. John Wiley & Sons 2012.
34. Kim, J. F.; Freitas da Silva, A. M.; Valtcheva, I. B.; Livingston, A. G., When the membrane is not
enough: A simplified membrane cascade using Organic Solvent Nanofiltration (OSN). Sep. Purif. Technol.
2013, 116, 277-286.
35. Kim, J. F.; Szekely, G.; Valtcheva, I. B.; Livingston, A. G., Increasing the sustainability of membrane
processes through cascade approach and solvent recovery-pharmaceutical purification case study. Green
Chem. 2014, 16, 133.
36. Székely, G.; Gil, M.; Sellergren, B.; Heggie, W.; Ferreira, F. C., Environmental and economic
analysis for selection and engineering sustainable API degenotoxification processes. Green Chem. 2013, 15,
210-225.
37. Xie, C.; Staszak, M.; Quatroche, J.; Sturgill, C.; Khau, V.; Martinelli, M., Nucleosidic
Phosphoramidite Synthesis via Phosphitylation: Activator Selection and Process Development. Org.
Process Res. Dev 2005, 9, 730-737.
38. Sanghvi, Y. S.; Guo, Z.; Pfundheller, H. M.; Converso, A., Improved process for the preparation of
nucleosidic phosphoramidites using a safer and cheaper activator. Org. Process Res. Dev. 2000, 4, 175-181.
39. Kim, J. F.; Szekely, G.; Schaepertoens, M.; Valtcheva, I. B.; Jimenez-Solomon, M. F.; Livingston, A.
G., In Situ Solvent Recovery by Organic Solvent Nanofiltration. ACS Sustainable Chemistry &
Engineering 2014, 2, 2371-2379.
40. Rundquist, E. M.; Pink, C. J.; Livingston, A. G., Organic solvent nanofiltration: a potential alternative
to distillation for solvent recovery from crystallisation mother liquors. Green Chem. 2012, 14, 2197.
41. Székely, G.; Schaepertoens, M.; Gaffney, P. R. J.; Livingston, A. G., Beyond PEG2000: Synthesis
and Functionalisation of Monodisperse PEGylated Homostars and Clickable Bivalent Polyethyleneglycols.
Chemistry – A European Journal 2014, 20, 10038-10051.
42. See-Toh, Y. H.; Silva, M.; Livingston, A., Controlling molecular weight cut-off curves for highly
solvent stable organic solvent nanofiltration (OSN) membranes. J. Membr. Sci. 2008, 324, 220-232.
43. Molina, A. G.; Kungurtsev, V.; Virta, P.; Lonnberg, H., Acetylated and Methylated β-Cyclodextrins
as Viable Soluble Supports for the Synthesis of Short 2'-Oligodeoxyribo-nucleotides in Solution. Molecules
2012, 17, 12102-12120.
44. Molina, A. G.; Jabgunde, A. M.; Virta, P.; Lönnberg, H., Solution phase synthesis of short
oligoribonucleotides on a precipitative tetrapodal support. Beilstein J. Org. Chem. 2014, 10, 2279-2285.
43 | P a g e
top related